This project will use theory and a theory-driven randomized microfinance experiment to measure the importance of precautionary motives relative to entrepreneurial / self-financing motives in driving the savings behavior of the poor in developing countries. This question is important because the two motives potentially suggest different predictions for household and macroeconomic responses to savings and credit interventions, or financial deepening more broadly. This project will develop a parsimonious model of household savings, consumption, and investment in the face of income shocks, entrepreneurial opportunities, and credit constraints. A simple parameterization will allow us to shift motives for saving from buffering income shocks to financing entrepreneurial investments. The model helps to motivate a simple experimental test: does increased access to in-kind investment loans vis-a-vis cash loans lead to differential impacts on households depending on the relative strength of these motives? In the model, the two types of loans will have similar impacts for self-financing savers (increasing investment, and consumption over time), but only the latter will impact precautionary savers (causing short-term increases in consumption). We will then implement this test using a randomized experiment with a microfinance organization in Uganda to measure the actual savings motives of the local population. We will first offer savings accounts to 3,000 households, which are members of the unbanked poor. Of these, one third will later be offered cash loans, one third will be offered in-kind loans under the same terms, and one third will be kept as a control. Three rounds of surveys will measure the households' behavior upon receiving the savings accounts, soon after receiving access to their loans, and then twelve months later. Finally, we will evaluate the results of the experiment, its implications for key parameters in the quantitative model, and the model's implications for alternative saving and lending scenarios / policies under those parameters.